1-samples_selection.py

from Packages import *
st.set_page_config(page_title="NIRS Utils", page_icon=":goat:", layout="wide")
from Modules import *
from Class_Mod.DATA_HANDLING import *



# HTML pour le bandeau "CEFE - CNRS"
bandeau_html = """
<div style="width: 100%; background-color: #4682B4; padding: 10px; margin-bottom: 10px;">
  <h1 style="text-align: center; color: white;">CEFE - CNRS</h1>
</div>
"""
# Injecter le code HTML du bandeau
st.markdown(bandeau_html, unsafe_allow_html=True)

st.session_state["interface"] = st.session_state.get('interface')
if st.session_state["interface"] == 'simple':
    hide_pages("Predictions")

################################### Data Loading and Visualization ########################################
col2, col1 = st.columns([3, 1])
col1.header("Data Loading", divider='blue')
col2.header("Spectral Data Visualization", divider='blue')


## Preallocation of data structure
spectra = pd.DataFrame
meta_data = pd.DataFrame
selected_samples = pd.DataFrame


# loader for datafile
data_file = col1.file_uploader("Load NIRS Data", type=["csv","dx"], help=" :mushroom: select a csv matrix with samples as rows and lambdas as columns", key=5)


if data_file:
    # Retrieve the extension of the file
    test = data_file.name[data_file.name.find('.'):]

    ## Load .csv file
    if test== '.csv':
        with col1:
            # Select list for CSV delimiter
            psep = st.radio("Select csv separator - _detected_: " + str(find_delimiter('data/'+data_file.name)), options=[";", ","], index=[";", ","].index(str(find_delimiter('data/'+data_file.name))), key=9)
                # Select list for CSV header True / False
            phdr = st.radio("indexes column in csv? - _detected_: " + str(find_col_index('data/'+data_file.name)), options=["no", "yes"], index=["no", "yes"].index(str(find_col_index('data/'+data_file.name))), key=31)
            if phdr == 'yes':
                col = 0
            else:
                col = False
            imp = pd.read_csv(data_file, sep=psep, index_col=col)
            # spectra = col_cat(imp)[0]
            # meta_data = col_cat(imp)[1]
            spectra, meta_data = col_cat(imp)
            st.success("The data have been loaded successfully", icon="✅")

    ## Load .dx file
    elif test == '.dx':
        # Create a temporary file to save the uploaded file
        with NamedTemporaryFile(delete=False, suffix=".dx") as tmp:
            tmp.write(data_file.read())
            tmp_path = tmp.name
            with col1:
                _, spectra, meta_data = read_dx(file = tmp_path)
                st.success("The data have been loaded successfully", icon="✅")
        os.unlink(tmp_path)

## Visualize spectra
if not spectra.empty:
    with col2:
        fig = plot_spectra(spectra)
        st.pyplot(fig)
        fig.savefig("./Report/figures/Spectra_Plot.png")

############################## Exploratory data analysis ###############################
container2 = st.container(border=True)
container2.header("Exploratory Data Analysis-Multivariable Data Analysis", divider='blue')
scores, loadings, pc = st.columns([2, 3, 0.5])
influence, hotelling, qexp = st.columns([2, 2, 1])
st.header('Selected samples for chemical analysis')
selected_s, selected_samples_metd = st.columns([3, 3])
selected_s.write('Samples scores')

dim_red_methods=['', 'PCA','UMAP', 'NMF']  # List of dimensionality reduction algos
cluster_methods = ['', 'Kmeans','HDBSCAN', 'AP'] # List of clustering algos

dr_model = None # dimensionality reduction model
cl_model = None # clustering model

# Dimensionality reduction
t = pd.DataFrame # scores
p = pd.DataFrame # loadings
labels = []
if not spectra.empty:
    dim_red_method = pc.selectbox("Dimensionality reduction techniques: ", options = dim_red_methods, key = 37)
    clus_method = pc.selectbox("Clustering techniques: ", options = cluster_methods, key = 38)
    xc = standardize(spectra)


    if dim_red_method == dim_red_methods[1]:
        dr_model = LinearPCA(xc, Ncomp=5)
    elif dim_red_method == dim_red_methods[2]:
        if not meta_data.empty:
            filter = meta_data.columns[1:]
            col = pc.selectbox('Supervised UMAP by:', options= filter, key=108)
            supervised = meta_data[col]
        else:
            supervised = None
        dr_model = Umap(data_import = imp, numerical_data = MinMaxScale(spectra), cat_data = supervised)

    if dr_model:
        axis1 = pc.selectbox("x-axis", options = dr_model.scores_.columns, index=0)
        axis2 = pc.selectbox("y-axis", options = dr_model.scores_.columns, index=1)
        axis3 = pc.selectbox("z-axis", options = dr_model.scores_.columns, index=2)
        t = pd.concat([dr_model.scores_.loc[:,axis1], dr_model.scores_.loc[:,axis2], dr_model.scores_.loc[:,axis3]], axis = 1)


# clustering
if not t.empty:
    tcr = standardize(t)
        # Clustering
    if clus_method == cluster_methods[1]:
        ncluster = scores.number_input(min_value=2, max_value=30, value=3, label = 'Select the desired number of clusters')
        cl_model = Sk_Kmeans(tcr, max_clusters = 30)
        fig2 = px.scatter(cl_model.inertia_.T, y = 'inertia')
        scores.plotly_chart(fig2,use_container_width=True)
        img = pio.to_image(fig2, format="png")
        with open("./Report/figures/Elbow.png", "wb") as f:
                f.write(img)    
        data, labels, clu_centers = cl_model.fit_optimal(nclusters = ncluster)

    elif clus_method == cluster_methods[2]:
        optimized_hdbscan = Hdbscan(dr_model.scores_raw_)
        labels, hdbscan_score = optimized_hdbscan.HDBSCAN_scores_

##### Plots


#####################################################################################################
selec_strategy = ['center','random']
samples_df_chem = pd.DataFrame
selected_samples = []
selected_samples_idx = []

if labels:
    selection = scores.radio('Select samples selection strategy:', options = selec_strategy)
#################### selection strategy to  be corrected
    if selection == selec_strategy[0]:
        # list samples at clusters centers - Use sklearn.metrics.pairwise_distances_argmin if you want more than 1 sample per cluster
        closest, _ = pairwise_distances_argmin_min(clu_centers, tcr)
        selected_samples_idx = list(closest)
    elif selection == selec_strategy[1]:
        selection_number = scores.number_input('How many samples per cluster?', min_value = 1, step=1, value = 3)
        for i in np.unique(labels):
            C = np.where(np.array(labels) ==i)[0]
            if C.shape[0] >= selection_number:
                #scores.write(list(tcr.index)[labels== i])
                km2 = KMeans(n_clusters = selection_number)
                km2.fit(tcr.iloc[C,:])
                clos, _ = pairwise_distances_argmin_min(km2.cluster_centers_, tcr.iloc[C,:])
                selected_samples_idx2 = list(clos)
                selected_samples_idx.extend(tcr.iloc[C,:].index[selected_samples_idx2])                
            #    selected_samples_idx.extend(tcr.iloc[C,:].sample(n=selection_number).index.to_list())
            else:
                selected_samples_idx.extend(tcr.iloc[C,:].index.to_list())
            # list indexes of selected samples for colored plot    

if labels:
    if selected_samples_idx:
        sam = pd.DataFrame({'cluster':np.array(labels)[selected_samples_idx],
                            'index': spectra.index[selected_samples_idx]})
        selected_s.write(sam)

        if not meta_data.empty:
            selected_samples_metd.write('Corresponding meta-data')
            meta = meta_data.iloc[selected_samples_idx,:]
            meta['cluster'] = np.array(labels)[selected_samples_idx]
            meta['index'] = spectra.index[selected_samples_idx]
            selected_samples_metd.write(meta)


    ## Scores
if not t.empty:
    with scores:
        st.write('Scores plot')
        # scores plot with clustering
        if list(labels) and meta_data.empty:
            fig = px.scatter_3d(tcr, x=axis1, y=axis2, z = axis3, color = labels)
    
        # scores plot with metadata
        elif len(list(labels)) == 0 and not meta_data.empty:
            filter = meta_data.columns[1:]
            col = st.selectbox('Color by:', options= filter)
            if col == 0:
                fig = px.scatter_3d(tcr, x=axis1, y=axis2, z = axis3)
            else:
                fig = px.scatter_3d(tcr, x=axis1, y=axis2, z = axis3, color = list(map(str.lower,meta_data[col])) )

        # color with scores and metadata
        elif len(list(labels)) > 0  and not meta_data.empty:
            if clus_method in cluster_methods[1:]:
                filter = ['None', clus_method]
                filter.extend(meta_data.columns[1:])
            else:
                filter = meta_data.columns[1:].insert(0,'None')

            col = st.selectbox('Color by:', options= filter)
            if col == "None":
                fig = px.scatter_3d(tcr, x=axis1, y=axis2, z = axis3)
            elif col == clus_method:
                fig = px.scatter_3d(tcr, x=axis1, y=axis2, z = axis3, color = labels)
            else:
                fig = px.scatter_3d(tcr, x=axis1, y=axis2, z = axis3, color = list(map(str.lower,meta_data[col])))

        else:
            fig = px.scatter_3d(tcr, x=axis1, y=axis2, z = axis3)
        fig.update_traces(marker=dict(size=4))

        if selected_samples_idx:
            tt = tcr.iloc[selected_samples_idx,:]
            fig.add_scatter3d(x = tt.loc[:,axis1], y = tt.loc[:,axis2],
                              z = tt.loc[:,axis3], mode ='markers', marker = dict(size = 7, color = 'black'),
                              name = 'selected samples')
        st.plotly_chart(fig, use_container_width=True)

        import plotly.express as px

        if labels:
            num_clusters = len(np.unique(labels))

            custom_color_palette = px.colors.qualitative.Plotly[:num_clusters]
            color_discrete_sequence=custom_color_palette

            # Créer et exporter le graphique Axe1-Axe2 en PNG
            fig_axe1_axe2 = px.scatter(tcr, x=axis1, y=axis2, color=labels if list(labels) else None, color_discrete_sequence=custom_color_palette)
            fig_axe1_axe2.update_layout(title='Axe1-Axe2')
            fig_axe1_axe2.update_traces(marker=dict(size=4))
            fig_axe1_axe2.write_image("./Report/Figures/plot_axe1_axe2.png")


            # Créer et exporter le graphique Axe1-Axe3 en PNG
            fig_axe1_axe3 = px.scatter(tcr, x=axis1, y=axis3, color=labels if list(labels) else None, color_discrete_sequence=custom_color_palette)
            fig_axe1_axe3.update_layout(title='Axe1-Axe3')
            fig_axe1_axe3.update_traces(marker=dict(size=4))
            fig_axe1_axe3.write_image("./Report/Figures/plot_axe1_axe3.png")


            # Créer et exporter le graphique Axe2-Axe3 en PNG
            fig_axe2_axe3 = px.scatter(tcr, x=axis2, y=axis3, color=labels if list(labels) else None, color_discrete_sequence=custom_color_palette)
            fig_axe2_axe3.update_layout(title='Axe2-Axe3')
            fig_axe2_axe3.update_traces(marker=dict(size=4))
            fig_axe2_axe3.write_image("./Report/Figures/plot_axe2_axe3.png")


if not spectra.empty:
    if dim_red_method == dim_red_methods[1]:

        with loadings:
            st.write('Loadings plot')
            p = dr_model.loadings_
            pp = pd.concat([p, pd.DataFrame(np.arange(p.shape[0]), index=p.index, columns=['wl'])], axis=1)
            df1 = pp.melt(id_vars="wl")
            fig = px.line(df1, x='wl', y='value', color='variable', color_discrete_sequence=px.colors.qualitative.Plotly)
            fig.update_layout(legend=dict(x=1, y=0, font=dict(family="Courier", size=12, color="black"),
                                        bordercolor="black", borderwidth=2))
            st.plotly_chart(fig, use_container_width=True)

            # Export du graphique
            img = pio.to_image(fig, format="png")
            with open("./Report/figures/graphe_loadings.png", "wb") as f:
                f.write(img)

        with influence:
            st.write('Influence plot')
            ax1 = st.selectbox("Component", options=dr_model.scores_.columns, index=3)
            leverage = dr_model.leverage_
            residuals = dr_model.residuals_
            fig = px.scatter(x=leverage[ax1], y=residuals[ax1], color=leverage[ax1]*residuals[ax1], color_continuous_scale='Blues')
            fig.update_layout(xaxis_title="Leverage", yaxis_title="Residuals")
            st.plotly_chart(fig)
            img = pio.to_image(fig, format="png")
            with open("./Report/figures/graphe_influence.png", "wb") as f:
                f.write(img)


        with hotelling:
            st.write('T²-Hotelling vs Q residuals plot')
            hotelling = dr_model.hotelling_
            ax2 = st.selectbox("Component", options=dr_model.scores_.columns, index=4)

            hotelling = dr_model.hotelling_
            fig = px.scatter(t, x=hotelling[ax2], y=residuals[ax2]).update_layout(xaxis_title="T²",yaxis_title="Residuals")
            st.plotly_chart(fig, use_container_width=True)
            fig.write_image("./Report/figures/graphe_hotelling.png", format="png")

    if dim_red_method == dim_red_methods[2] and clus_method == cluster_methods[2]: # UMAP clustered by HDBSCAN
        with loadings: # Display some clustering metrics
            st.write('Clustering metrics:')
            clusters_number = set(labels)
            clusters_number.remove(-1)
            st.write('Optimal number of clusters = ' + str(len(clusters_number)))
            st.write('DBCV score (-1 to 1 - higher is better) = ' + str(round(hdbscan_score,3)))
            st.write('Unclassified samples: ' + str(len(t[labels==-1])) + ' on ' + str(len(t)) + ' samples (' + str(round(len(t[labels==-1])/len(t)*100, 1)) + '%).')